Assessing the transferability of support vector machine model for estimation of global solar radiation from air temperature

Exploring novel methods for estimation of global solar radiation from air temperature has been being a focus in many studies. This paper evaluates the transferability of support vector machines (SVM) for estimation of solar radiation in subtropical zone in China. Results suggest that solar radiation at one site (estimation site) could be well estimated by SVM model developed at another site (source site). The accuracy of estimation is affected by the distance and temperature difference between two sites, and altitude of source site. Higher correlations between RMSE of SVM and distance, and temperature differences are observed in northeastern region, increasing the reliability and confidence of SVM model developed at nearby stations. While lower correlations between RMSE and distance, and temperature differences are observed in southwest plateau region. When the altitude of estimation site is lower than 1200 m, RMSE show logarithm relationship with altitude of source sites where the altitude are lower than that of estimation site. Otherwise, RMSE show linearly relationship with altitude of source sites where the altitude are higher than 200 m but lower than that of the estimation site. This result suggests that solar radiation could be also estimated using SVM model developed at the site with similar but lower altitude. Based on these results, a strategy that takes into account the climatic conditions, topography, distance, and altitude for selecting a suitable source site is presented. The findings can guide and ease the appropriate choice of source sites for estimation of solar radiation at estimation site.